Forecasting Financial Returns Volatility Based On Some Classes Of GARCH-SVR Models

Volatility is one of the most important characteristics of financial time series and one of the important influencing factors for personal investment,corporate decision-making,and government regulation.The volatility of financial time series is characterized by structural breaks,asymmetry,and long memory.In recent years,computer technology has developed rapidly,and machine learning methods have also been widely applied,especially in dealing with nonlinear problems,which have obvious advantages and can provide more accurate predictions.In this paper,based on two classes of GARCH models,combined with the support vector regression(SVR)model in machine learning,a hybrid model is obtained,and studying the forecast performance of this model on volatility.The main work of this paper is described in detail below.In the first part of this paper,in order to capture the structural breaks of financial time series volatilitys,we propose a Markov-switching GARCH SVR(MS-GARCH-SVR)model based on the combination of the classical Markov-switching GARCH(MS-GARCH)model and the SVR model.At the same time,in order to characterize the heavy tail characteristics of financial time series,we consider two different distributions of normal distribution and distribution.Through simulation studies,the forecast performance of MS-GARCH-SVR model are compared with MS-GARCH model and mainstream hybrid model respectively.The simulation results show that the forecast performance of all hybrid models is improved compared with MS-GARCH model.Secondly,compared with mainstream hybrid models,the MS-GARCH-SVR model has the smallest MSE and MAE values,It is proved that the hybrid method proposed in this paper has the best forecast performance.In the empirical analysis,the MS-GARCH-SVR model is applied to the data analysis of S&P 500 index,EURUSD and BPAmoco.The MS-GARCH-SVR model can well characterize the structural breaks and heavy-tailed characteristics of volatilitys.Considering that the MS-GARCH model ignores the long memory characteristic of some financial time series volatilitys,as well as the "pseudo long memory" situation that often occurs in financial time series due to structural breaks.In the second part of this paper,we further studied models that can characterize structural breaks and long memory characteristics.We considered combining the Markov-switching fractionally integrated GARCH(MS-FIGARCH)model and the SVR model to obtain the MS-FIGARCH-SVR model.By simulation studies,the forecast performance of the MS-FIGARCH-SVR model was compared with the MS-FIGARCH model and hybrid models based on MS-FIGARCH model,respectively.Simulation studies have shown that compared to the MS-FIGARCH model,all hybrid models have improved prediction accuracy.Secondly,compared with hybrid models,the MS-FIGARCH-SVR model has smaller MSE and MAE values,proving that the hybrid method proposed in this paper has good prediction accuracy.In empirical analysis,the MS-FIGARCH-SVR model was applied to the daily closing price data analysis of Brent crude oil in Europe(Brent),gold(Gold),and CSI300.The empirical analysis results showed that the MS-FIGARCH-SVR model can well characterize the long-term memory characteristics of volatility.